Re: [PATCH] futex: Fix potential use-after-free in FUTEX_REQUEUE_PI

[Darren Hart <dvhart@infradead.org>]

On Thu, Nov 24, 2016 at 04:38:08PM +0100, Peter Zijlstra wrote:
> On Thu, Nov 24, 2016 at 04:19:41PM +0100, Thomas Gleixner wrote:
> > On Thu, 24 Nov 2016, Peter Zijlstra wrote:
> > 
> > > 
> > > While working on the futex code, I stumbled over this potential
> > > use-after-free scenario.
> > > 
> > > pi_mutex is a pointer into pi_state, which we drop the reference on in
> > > unqueue_me_pi(). So any access to that pointer after that is bad.
> > > 
> > > Since other sites already do rt_mutex_unlock() with hb->lock held, see
> > > for example futex_lock_pi(), simply move the unlock before
> > > unqueue_me_pi().
> > > 
> > > Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
> > > ---
> > >  kernel/futex.c | 22 +++++++++++++---------
> > >  1 file changed, 13 insertions(+), 9 deletions(-)
> > > 
> > > diff --git a/kernel/futex.c b/kernel/futex.c
> > > index 2c4be467fecd..d5a81339209f 100644
> > > --- a/kernel/futex.c
> > > +++ b/kernel/futex.c
> > > @@ -2813,7 +2813,6 @@ static int futex_wait_requeue_pi(u32 __user *uaddr, unsigned int flags,
> > >  {
> > >  	struct hrtimer_sleeper timeout, *to = NULL;
> > >  	struct rt_mutex_waiter rt_waiter;
> > > -	struct rt_mutex *pi_mutex = NULL;
> > >  	struct futex_hash_bucket *hb;
> > >  	union futex_key key2 = FUTEX_KEY_INIT;
> > >  	struct futex_q q = futex_q_init;
> > > @@ -2905,6 +2904,8 @@ static int futex_wait_requeue_pi(u32 __user *uaddr, unsigned int flags,
> > >  			spin_unlock(q.lock_ptr);
> > 
> > In this path the fixup can return -EFAIL as well, so it should drop rtmutex
> > too if it owns it. We should move the rtmutex drop into the fixup functions...

I traced through the possible return codes and found:

fixup_pi_state_owner
  see below

rt_mutex_finish_proxy_lock
  __rt_mutex_slowlock
    EINTR
    ETIMEDOUT
(ignored if fixup_owner fails)

fixup_owner
  fixup_pi_state_owner
    fault_in_user_writeable
      fixup_user_fault
        EFAULT
        ENOMEM
        EHWPOISON

> 
> Urgh, so would really like to avoid doing that, I'll have to instantly
> drag it back out again :/
> 
> Also, the fixup_owner() fail in futex_lock_pi() will unlock the rt_mutex
> on _any_ fail, not only -EFAULT, should we not do the same? 
> 

I don't see why we should treat ENOMEM or EHWPOISON any differently from EFAULT
in this situation.

> ---
> Subject: futex: Fix potential use-after-free in FUTEX_REQUEUE_PI
> From: Peter Zijlstra <peterz@infradead.org>
> Date: Thu, 24 Nov 2016 15:42:35 +0100
> 
> While working on the futex code, I stumbled over this potential
> use-after-free scenario.
> 
> pi_mutex is a pointer into pi_state, which we drop the reference on in
> unqueue_me_pi(). So any access to that pointer after that is bad.
> 
> Since other sites already do rt_mutex_unlock() with hb->lock held, see
> for example futex_lock_pi(), simply move the unlock before
> unqueue_me_pi().
> 
> Cc: Ingo Molnar <mingo@kernel.org>
> Cc: dvhart@infradead.org
> Cc: Thomas Gleixner <tglx@linutronix.de>
> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
> ---
>  kernel/futex.c |   22 +++++++++++++---------
>  1 file changed, 13 insertions(+), 9 deletions(-)
> 
> --- a/kernel/futex.c
> +++ b/kernel/futex.c
> @@ -2813,7 +2813,6 @@ static int futex_wait_requeue_pi(u32 __u
>  {
>  	struct hrtimer_sleeper timeout, *to = NULL;
>  	struct rt_mutex_waiter rt_waiter;
> -	struct rt_mutex *pi_mutex = NULL;
>  	struct futex_hash_bucket *hb;
>  	union futex_key key2 = FUTEX_KEY_INIT;
>  	struct futex_q q = futex_q_init;
> @@ -2897,6 +2896,8 @@ static int futex_wait_requeue_pi(u32 __u
>  		if (q.pi_state && (q.pi_state->owner != current)) {
>  			spin_lock(q.lock_ptr);
>  			ret = fixup_pi_state_owner(uaddr2, &q, current);
> +			if (ret && rt_mutex_owner(&q.pi_state->pi_mutex) == current)
> +				rt_mutex_unlock(&q.pi_state->pi_mutex);
>  			/*
>  			 * Drop the reference to the pi state which
>  			 * the requeue_pi() code acquired for us.
> @@ -2905,6 +2906,8 @@ static int futex_wait_requeue_pi(u32 __u
>  			spin_unlock(q.lock_ptr);
>  		}
>  	} else {
> +		struct rt_mutex *pi_mutex;
> +
>  		/*
>  		 * We have been woken up by futex_unlock_pi(), a timeout, or a
>  		 * signal.  futex_unlock_pi() will not destroy the lock_ptr nor
> @@ -2928,18 +2931,19 @@ static int futex_wait_requeue_pi(u32 __u
>  		if (res)
>  			ret = (res < 0) ? res : 0;
>  
> +		/*
> +		 * If fixup_pi_state_owner() faulted and was unable to handle

faulted or failed ?

> +		 * the fault, unlock the rt_mutex and return the fault to

propagate the error to userspace

> +		 * userspace.
> +		 */
> +		if (ret && rt_mutex_owner(pi_mutex) == current)
> +			rt_mutex_unlock(pi_mutex);
> +
>  		/* Unqueue and drop the lock. */
>  		unqueue_me_pi(&q);
>  	}
>  
> -	/*
> -	 * If fixup_pi_state_owner() faulted and was unable to handle the
> -	 * fault, unlock the rt_mutex and return the fault to userspace.
> -	 */
> -	if (ret == -EFAULT) {
> -		if (pi_mutex && rt_mutex_owner(pi_mutex) == current)
> -			rt_mutex_unlock(pi_mutex);
> -	} else if (ret == -EINTR) {
> +	if (ret == -EINTR) {
>  		/*
>  		 * We've already been requeued, but cannot restart by calling
>  		 * futex_lock_pi() directly. We could restart this syscall, but
> 

-- 
Darren Hart
Intel Open Source Technology Center